[Metricbeat][Docker] Docker cpu norms

[ChrsMark]

This PR implements the proposal for docker.cpu.*.norm.pct fields described on #13343.

How to test it

1. Enable Docker Module and cpu metricset specifically.
2. Run Metricbeat and check that all the new normalised fields are present:

• kernel.norm.pct
• system.norm.pct
• user.norm.pct
• total.norm.pct
• core.*.norm.pct

[fearful-symmetry]

Hey, can we get Godocs for the rest of the exported methods in helper.go ? Surprised hound didn't complain.

[ChrsMark]

Hey, can we get Godocs for the rest of the exported methods in helper.go ? Surprised hound didn't complain.

👍 sure! thank you!

[jsoriano]

In principle apart of the small comment it LGTM.

I have a couple of questions before merging:

In the system modules, the cpu metrics collected are optional and managed by the cpu.metrics setting. It can be useful if more methods are added, though I think that not really needed because the overhead doesn't look so big. Did you consider doing something similar here?

Not to be done in this PR, but I think that it could be good to expose the data to calculate percentages in the events, so custom percentages can be calculated on query time. Now we have something like the number of cpus of the host in docker.cpu.system.pct, but it is a bit counterintuitive, should we add a explicit field for that? like docker.cpu.online_cpus?

[jsoriano]

Remove commented out code.

[ChrsMark]

🙈 thank you for catching this!

[jsoriano]

Also a changelog entry will be needed.

[ChrsMark]

In principle apart of the small comment it LGTM.

Thank you for reviewing this Jaime!

I have a couple of questions before merging:

In the system modules, the cpu metrics collected are optional and managed by the cpu.metrics setting. It can be useful if more methods are added, though I think that not really needed because the overhead doesn't look so big. Did you consider doing something similar here?

Not to be done in this PR, but I think that it could be good to expose the data to calculate percentages in the events, so custom percentages can be calculated on query time. Now we have something like the number of cpus of the host in docker.cpu.system.pct, but it is a bit counterintuitive, should we add a explicit field for that? like docker.cpu.online_cpus?

Yeap I was thinking about this too, however I didn't add this yet since I was not sure enough because it was not already there for pct and ticks.
I'm really interested in having this extension in this metriset too and I could work on this. We could open another issue to discuss the specs, WDYT?

[jsoriano]

Ok, let's go on with this and please open another issue to add metrics about limited resources. Apart of adding metrics about the available CPUs, we could also check if we are collecting metrics about available memory.